Ramanujan–Fourier series, the Wiener–Khintchine

Physica A 269 (1999) 503–510
www.elsevier.com/locate/physa
Ramanujan–Fourier series, the Wiener–Khintchine
formula and the distribution of prime pairs
H. Gopalkrishna Gadiyar, R. Padma ∗
Ramanujan Institute for Advanced Study in Mathematics, University of Madras, Chennai 600 005, India
Received 3 February 1999
Abstract
The Wiener–Khintchine formula plays a central role in statistical mechanics. It is shown here
that the problem of prime pairs is related to autocorrelation and hence to a Wiener–Khintchine
c 1999 Elsevier Science B.V. All rights
formula. “Experimental” evidence is given for this. reserved.
PACS: 05.40+j; 02.30.Nw; 02.10.Lh
Keywords: Twin primes; Ramanujan–Fourier series; Wiener–Khintchine formula
1. Introduction
“ The Wiener–Khintchine theorem states a relationship between two important characteristics of a random process: the power spectrum of the process and the correlation
function of the process” [1]. One of the outstanding problems in number theory is the
problem of prime pairs which asks how primes of the form p and p +h (where h is an
even integer) are distributed. One immediately notes that this is a problem of nding
correlation between primes. We make two key observations. First of all there is an
arithmetical function (a function dened on integers) which traps the properties of the
primes. This is the von Mangoldt function (n) which is dened to be equal to log p
if n = p k , where p is prime and k any positive integer and is equal to 0 otherwise.
This is a good approximate “weighted” characteristic function of primes. Recall that a
characteristic function f(x) is equal to 1 if x ∈ S and 0 if x ∈= S for some set S. It is
standard to use (n) in number theory in stead of characteristic function of primes.
∗
Corresponding author.
E-mail address: padma@imsc.ernet.in (R. Padma)
c 1999 Elsevier Science B.V. All rights reserved.
0378-4371/99/$ - see front matter PII: S 0 3 7 8 - 4 3 7 1 ( 9 9 ) 0 0 1 7 1 - 5
504
H.G. Gadiyar, R. Padma / Physica A 269 (1999) 503–510
The second observation is that such arithmetical functions have a Ramanujan–Fourier
series. This is due to Ramanujan. The papers of Ramanujan [2] and Hardy [3] are still
the best place to read this topic. In this paper we show that the problem of distribution of prime pairs is equivalent to proving a Wiener–Khintchine formula. Further,
numerical evidence is provided to justify the plausibility of this approach.
2. Ramanujan–Fourier series
The Ramanujan–Fourier series of an arithmetical function a(n) is an expansion of
the form
∞
X
aq cq (n) ;
(1)
a(n) =
q=1
where
cq (n) =
q
X
e2i(k=q)n
k=1
(k; q)=1
and (k; q) denotes the greatest common divisor of k and q. Using extremely simple
arguments Ramanujan showed that the commonly known arithmetical functions all have
Ramanujan–Fourier series. For example, he showed that
∞
X
log q
cq (n) ;
d(n) = −
q
q=1
(n) =
∞
n X cq (n)
;
6
q2
2
q=1
where d(n) is the number of divisors of n and (n) their sum. He however did not
indicate any formula for getting the Ramanujan–Fourier coecients aq which are the
back bone of Fourier analysis. This was done by Carmichael [4] a little later. He showed
that e2i(k=q)n are almost periodic functions dened on the integers. This led to the
orthogonality relations and a method for evaluating the Ramanujan–Fourier coecient.
Denote by M (g) the mean value of an arithmetical function g, that is,
1 X
g(n) :
M (g) = lim
N →∞ N
n6N
For 16k6q; (k; q)=1; let ek=q (n)=e2i(k=q)n ; (n ∈ N): If a(n) is an arithmetical function
with expansion (1) then
1 X
1
1
a(n)cq (n) :
M (acq ) =
lim
aq =
(q)
(q) N →∞ N n6N
Also,
(
M (ek=q ek 0 =q0 ) =
1
if k=q = k 0 =q0 ;
0 if k=q 6= k 0 =q0 :
All this is well known in the eld of almost periodic functions.
(2)
H.G. Gadiyar, R. Padma / Physica A 269 (1999) 503–510
505
The arithmetical properties of cq (n) were elucidated by Hardy in [3] using which
he obtained a Ramanujan–Fourier expansion for ((n)=n)(n). (Here (n) denotes the
number of integers less than n and relatively prime to n and is known as the Euler
totient function.) We state these results below.
(i) cq (n) is multiplicative. That is,
cqq0 (n) = cq (n)cq0 (n)
if (q; q0 ) = 1 :
(3)
(ii) If p is a prime, then
−1
if pAn ;
cp (n) =
p − 1 if p|n ;
(4)
where a | b means a divides b and aAb means a does not divide b.
(iii)
∞
X (q)
(n)
(n) =
cq (n) :
n
(q)
(5)
q=1
Here (q) is the Mobius function dened as follows:
if q = p1 p2 · · · pk ; pi0 s are distinct primes ;
(−1)k
(q) =
0
otherwise :
(6)
Also,
(iv) For a given integer h,
∞
X
2 (q)
cq (h)
2 (q)
q=1
Y  Y
1
p−1


1
−
2

(p − 1)2
p−2
p¿2
p|h
=

p¿2


0
def
C(h) =
Q
if h is even ;
(7)
if h is odd ;
where p denotes the product over primes.
The proof of (7) follows from the multiplicative properties of (q); (q) and cq .
For a given h, the series on the left-hand side of (7) is absolutely convergent and
hence has the Euler product expansion
Y
Y
Y
2 (p)
1
1
cp (h) =
;
(8)
1+ 2
1−
1
+
(p)
(p − 1)2
p−1
p
pAh
p|h
by (4) and (6). When h is odd, the innite product on the right-hand side of (8) is 0
and when h is even, it is equal to
Y −1
Y
1
1
1
1−
1−
1+
;
2
(p − 1)2
p−1
(p − 1)2
p¿2
p|h
p¿2
which on simplication gives the right-hand side of (7).
506
H.G. Gadiyar, R. Padma / Physica A 269 (1999) 503–510
3. The twin-prime conjecture and the Wiener–Khintchine formula
The Wiener–Khintchine formula basically says that if
X
fn ein t ;
f(t) =
n
then
1
lim
T →∞ 2T
Z
T
−T
f(t + )f(t) dt =
X
|fn |2 ein :
(9)
n
The left-hand side of (9) is called an autocorrelation function. The right-hand side is
nothing but the power spectrum. It is used practically to extract hidden periodicities in
seemingly random phenomena.
For an arithmetical function a(n) having the Ramanujan–Fourier series (1), the
Wiener–Khintchine formula can be stated as follows:
∞
X
1 X
a(n)a(n + h) =
a2q cq (h) :
(10)
lim
N →∞ N
n6N
q=1
The proof of (10) will follow from (1) and (2) if the Ramanujan–Fourier expansion
of a(n) is absolutely and uniformly convergent. But the Ramanujan–Fourier series for
((n)=n)(n) does not belong to this class of functions. However, assuming the truth
of the theorem in the most general case would give the formula conjectured by Hardy
and Littlewood [5] which we now state.
There are inÿnitely many prime pairs p; p+h for every even integer h and if h (N )
denotes the number of prime pairs less than N , then
h (N ) ∼ C(h)
N
:
log 2 N
(11)
The Wiener–Khintchine formula for ((n)=n)(n) is given by
∞
X
(n + h)
1 X (n)
2 (q)
(n)
(n + h) =
cq (h) = C(h) :
N →∞ N
n
n+h
2 (q)
n6N
lim
(12)
q=1
Let
(h; N ) =
X (n)
(n + h)
(n)
(n + h) :
n
n+h
n6N
Then (12) implies that
(h; N )
∼ C(h) :
(13)
N
The terms of the sum on the left-hand side of (13) are non zero if and only if both
(n) and (n + h) are prime powers, say p k and ql , respectively, for primes p and
q. Then (n)=n = (p − 1=p) and (n + h)=(n + h) = (q − 1)=q. Hence,
X (n)
X
(n + h)
(n)(n + h) −
(n)
(n + h)
n
n+h
n6N
n6N
H.G. Gadiyar, R. Padma / Physica A 269 (1999) 503–510
=
X
p k 6N
p k +h=ql
log p log q
+
p
X
p k 6N
p k +h=ql
log p log q
−
q
X
p k 6N
p k +h=ql
507
log plog q
pq
=O(log 2 N log log N ) ;
(14)
where we have used the formula [6]
X log p
∼ log x
p
p6x
at the last step. Thus if (12) is true and h is even, then from (13) and (14) we get
X
(n)(n + h) ∼ C(h)N :
(15)
n6N
When we pass from this formula to a formula for h (N ) using partial summation, (see
for example, Section 3 of [7]), the formula which arises naturally is not (11) but
Z N
dx
(16)
h (N ) ∼ C(h)
2
2 log x
and this is naturally equivalent to (11).
When h is odd, we know from (7) that C(h) = 0. If a term on the left-hand side of
(13) is non-zero, then either n or n + h is a power of 2 and hence
X
X
(n)(n + h) = log 2
log p
n6N
2k 6N
|pl −2k |=h
= O(log 2 N )
(17)
from which one can deduce that h (N ) = O(1) when h is odd.
It is well-known that one proves the prime number theorem
X
x
1∼
(x) =
log x
p6x
by proving an equivalent assertion
X
(n) ∼ x ;
(x) =
(18)
n6x
for the Chebyshev function
this.
(x). Note that passing from (15) to (11) is analogous to
4. Numerical evidence
For completeness we give the compelling numerical evidence for the truth of the
Wiener–Khintchine formula. For h = 2; 4 and 6, the formula (13) gives
(2; N ) ∼ C(2)N ;
(4; N ) ∼ C(4)N ;
(6; N ) ∼ C(6)N :
508
H.G. Gadiyar, R. Padma / Physica A 269 (1999) 503–510
Table 1
N
(2; N )
(2; N )
N
Ratio
100000
200000
300000
400000
500000
600000
700000
800000
900000
1000000
131522.552204
264287.347531
393317.025988
525523.270611
654557.716460
789035.163302
919941.157912
1049182.174335
1180813.946552
1312843.985016
1.315226
1.321437
1.311057
1.313808
1.309115
1.315059
1.314202
1.311478
1.312015
1.312844
1.003876
0.999158
1.007068
1.004959
1.008562
1.004004
1.004658
1.006745
1.006332
1.005697
N
(4; N )
(4; N )
N
Ratio
100000
200000
300000
400000
500000
600000
700000
800000
900000
1000000
130212.335085
260492.247225
390320.617781
527155.226011
653649.051733
789177.513123
923982.224287
1054670.388142
1180133.117590
1307978.775955
1.302123
1.302461
1.301069
1.317888
1.307298
1.315296
1.319975
1.318338
1.311259
1.307979
1.013977
1.013714
1.014799
1.001848
1.009964
1.003823
1.000264
1.001506
1.006913
1.009438
Table 2
Note that C(4) = C(2) and C(6) = 2C(2). It follows that there should be approximately
equal numbers of prime power pairs diering by 2 and by 4, but about twice as many
diering by 6. We have tabulated below the actual values of (h; N ) and also the
ratio C(h)=((h; N )=N ) (third column) for h = 2; 4 and 6 and N upto 106 (see Tables
Q
1–3). We have used the value of C(2) = 2 p¿2 (1 − 1=(p − 1)2 ) ∼ 1:320323632 to
compute the ratio.
This shows that there is remarkable agreement between theory and numerical experiment. This is impressive evidence for the truth of the Wiener–Khintchine formula.
5. Conclusion
As Kac [8] remarks “..... Consider the integers divisible by both p and q (q another prime). To be divisible by p and q is equivalent to being divisible by pq and
consequently the density of the new set is 1=pq. Now,
11
1
;
=
pq p q
H.G. Gadiyar, R. Padma / Physica A 269 (1999) 503–510
509
Table 3
N
(6; N )
(6; N )
N
Ratio
100000
200000
300000
400000
500000
600000
700000
800000
900000
1000000
261289.742091
523391.109218
787393.641752
1056087.319082
1316336.875799
1579274.310330
1839327.388416
2104826.034045
2368450.398104
2631198.406265
2.612897
2.616956
2.624645
2.640218
2.632674
2.632124
2.627611
2.631033
2.631612
2.631198
1.010620
1.009053
1.006097
1.000162
1.003029
1.003238
1.004961
1.003654
1.003434
1.003591
and we can interpret this by saying that the “events” of being divisible by p and q are
independent. This holds, of course, for any number of primes, and we can say using
a picturesque but not a very precise language, that the primes play a game of chance!
This simple, nearly trivial, observation is the beginning of a new development which
links in a signicant way number theory on the one hand and probability theory on
the other”. One can see the interesting article [9] for elaborations of this theme. For
recent work in this direction see [10 –12].
The Wiener–Khintchine formula seems to show that the primes have structure hidden
in their seemingly arbitrary behaviour. As Hardy remarks in [3], “... These series have
a peculiar interest because they show explicitly the source of the irregularities in the
behaviour of their sums. Thus, for example, ...
(−1)n
2 cos(2=3)n 2 cos(1=2)n
2 n
1+
+
+
(n) =
6
22
32
42
2(cos(2=5)n + cos (4=5)n) 2 cos(1=3)n
+
+
+ ···
52
62
and we see at once that the most important term in (n) is 16 2 n, and that irregular
variations about this average value are produced by a series of harmonic oscillations of
decreasing amplitude”. Hence the Ramanujan–Fourier series trap the vagaries in the behaviour of primes and the Wiener–Khintchine formula traps their correlation properties.
It is a pleasant surprise that the Wiener–Khintchine formula which normally occurs in
practical problems of brownian motion, electrical engineering and other applied areas
of technology and statistical physics has a role in the behaviour of prime numbers
which are studied by pure mathematicians.
Acknowledgements
The second author wishes to thank the CSIR for nancial support.
510
H.G. Gadiyar, R. Padma / Physica A 269 (1999) 503–510
References
[1] C. Kittel, Elementary Statistical Physics, Wiley, New York, 1958.
[2] S. Ramanujan, On certain trigonometrical sums and their applications in the theory of numbers, Trans.
Camb. Phil. Soc. 22 (1918) 259.
[3] G.H. Hardy, Note on Ramanujan’s trigonometrical function cq (n) and certain series of arithmetical
functions, Proc. Camb. Phil. Soc. 20 (1921) 263.
[4] R.D. Carmichael, Expansions of arithmetical functions in innite series, Proc. London Math. Soc. 34
(2) (1932) 1.
[5] G.H. Hardy, J.E. Littlewood, Some problems of Partition Numerorum; III: On the expression of a
number as a sum of primes, Acta Math. 44 (1922) 1.
[6] T.M. Apostol, Introduction to Analytic Number Theory, International Springer Student Edition, Springer,
Berlin, 1989.
[7] H.G. Diamond, Elementary methods in the study of the distribution of prime numbers, Bull. (New
Series) Amer. Math. Soc. 7 (1982) 553.
[8] M. Kac, Statistical Independence in Probability, Analysis and Number Theory, The Carus Mathematical
Monograph No. 12, The Mathematical Association of America, 1964.
[9] P. Billingsley, Prime numbers and Brownian motion, Amer. Math. Monthly 80 (1973) 1099.
[10] M. Wolf, 1=f noise in the distribution of prime numbers, Physica A 241 (1997) 493.
[11] M. Wolf, Random walk on the prime numbers, Physica A 250 (1998) 335.
[12] N.M. Katz, P. Sarnak, Zeros of zeta functions and symmetry, Bull. (New Series) Amer. Math. Soc. 36
(1999) 1.